Research → Product → Code

The handoff is where research dies

Most products move an idea through a relay. A researcher frames the question and hands findings to a designer. The designer interprets them and hands mockups to an engineer. The engineer builds what they understood. Each pass is a lossy compression: the caveat a researcher fought for, the edge case, the reason behind a recommendation. A little of it is flattened at every step, until what ships is a confident artifact only loosely connected to what was actually learned. Nobody is at fault. The structure leaks.

One loop, not three relays

The alternative I keep returning to is to collapse the relay into a loop: the same person, or a tight pair, carries an idea from research, through design, into production code, and back again. This has nothing to do with one generalist beating three specialists. They usually don’t on any single axis. It works because the translation losses disappear when there is nothing to translate. When the person who ran the study also writes the component, the caveat survives, because it never had to be re-explained to anyone.

The handoff isn’t a step in the process. It’s where the process leaks.

What you actually gain

• Tighter feedback loops. A research insight becomes a working, testable prototype in hours, and the result feeds the next question the same day, not the next quarter.
• Judgment in the right place. Decisions about what to build are made by someone holding the evidence and the implementation constraints in the same head, at the same time.
• Defensible artifacts. The shipped thing carries its own reasoning, because the reasoning never had to leave the room.

At Articos I authored the research, designed the experience, and wrote the production code, and the rubric that defined a good report was the same rubric that gated the pipeline. Research and code were two views of one decision. At AutoLeap the loop ran tighter than the org chart: the inspection workflow I researched was the one I prototyped to high fidelity and shipped.

What it costs, and when it doesn’t apply

This is not a universal prescription, and pretending it is sets people up to fail. The loop trades depth-in-one-craft for range; a specialist will out-design or out-engineer a generalist on any single axis. It doesn’t scale linearly. You can’t staff a hundred-person org with people who do all three well, and the range is genuinely rare and hard-won. Where it earns its keep is 0→1 work and high-ambiguity problems: exactly where handoff loss is most expensive and speed of learning matters most. On a mature, well-specified surface, relays are fine. Specialize and hand off.

The range is the method

So the operating model isn’t “one person does everything.” It’s narrower and more useful than that: keep the distance between a question and a shipped answer as short as possible, and put judgment where the evidence and the constraints meet. When I can hold research, design, and code in one head, I do. The reason is not to prove I can. It is that fewer handoffs means fewer places for the work to leak. The range is the method, and that is the whole claim.

Further reading

For the design philosophy underneath this, AI as a Design Material sets out how I design with AI as a probabilistic raw material. And Auditable AI Research shows what the loop produces when the output has to be defended, not just read.